Key takeaways
- A WhatsApp chatbot runs on the WhatsApp Business API and answers customers automatically, either as a scripted menu or, with AI, in natural language.
- AI agents differ from rule-based bots in that they understand free text and answer from a knowledge base, but they must be grounded in your content to stay accurate.
- The handoff to a human is a design decision: define when, to whom and with what summary, and the bot becomes a first line rather than a wall.
- Measure containment rate, handoff rate, response time and satisfaction, and refine the knowledge base where the bot escalates most.
What a WhatsApp chatbot is, and what it is not
A WhatsApp chatbot is software that answers messages on a business WhatsApp number automatically. It runs on the WhatsApp Business API, not on the consumer app, and it can greet customers, answer questions, collect information, send documents and trigger actions in other systems, at any hour and in any volume.
It is not a replacement for the team. The best deployments treat the bot as the first line: it resolves the routine majority instantly and passes the rest to a person with everything already gathered. A bot that traps customers in loops or refuses to escalate damages the brand faster than a slow reply.
Rule-based bots vs AI agents
Two generations of WhatsApp bots are in use today, and many businesses combine them.
| Rule-based bot (flow) | AI agent | |
|---|---|---|
| How it understands | Keywords, buttons, numbered menus | Natural language, in any phrasing |
| Where answers come from | Scripted messages you write | Your knowledge base, documents and website |
| Best for | Fixed processes: opening hours, order status, booking steps | Open questions, product advice, troubleshooting, qualification |
| Risk | Rigid; fails on unexpected input | Can guess if not grounded in your content |
| Setup | Design the flow branch by branch | Describe the role, upload content, set rules |
| Handoff | On a menu option | On intent, sentiment, confidence or a rule you define |
In Teloring both exist side by side: Studio builds rule-based flows with triggers, conditions and actions, while AI Agents handle free conversation. A flow can hand a conversation to an AI agent, and the agent can hand it to a person.
What a WhatsApp AI bot can do for your business
- Answer the frequent questions instantly. Prices, availability, opening hours, policies and how-to steps, taken from your own content and with a reference to the source.
- Qualify leads. Ask what the customer needs, their budget, timing and location, write the answers into CRM fields, and route hot leads to sales.
- Book, confirm and remind. Collect appointment details and trigger the confirmation and reminder templates automatically.
- Collect structured data. Order numbers, addresses, ID numbers and photos, before a human ever opens the conversation.
- Speak the customer's language. Detect the language of the message and answer in it, in the tone you define.
- Work outside business hours. Cover nights and weekends, and let the morning team start with a summary rather than a backlog.
- Hand off with context. Pass the conversation to the right team with a summary, the collected fields and a suggested next step.
Keeping the bot honest: knowledge grounding
The risk with any AI chatbot is confident invention. The remedy is grounding: the agent answers only from content you have approved, such as PDFs, Word and Excel files, web pages you point it at or articles you write, and says so when it does not know. A well-grounded agent cites the document it drew from, so the team can check and improve the source.
Grounding also makes maintenance simple. When a price, policy or process changes, you update the knowledge base once, and every channel (WhatsApp, website chat, Instagram) answers correctly from that moment. In Teloring, the Knowledge Base accepts uploaded files and crawled web pages, and each AI Agent can be set to answer strictly from it, with a confidence threshold below which it hands off.
Handing off to a human
Decide the handoff rules before launch. Typical triggers are: the customer asks for a person; the agent's confidence is low; the topic is sensitive (complaints, cancellations, payments); sentiment turns negative; or a required field cannot be collected. Define where each case goes (a team, a specific agent, a queue) and what the person receives: the full transcript, a summary, the fields collected and the customer's record.
Working hours matter too. Outside them, the agent can collect details and set expectations for a reply; inside them, it can hand off live. When the conversation returns to the agent, say after the person resolves the issue and closes the chat, the next message should start fresh with the same knowledge, not the same loop.
Setting up an AI agent on WhatsApp step by step
- 1. Connect WhatsApp through the Business API. The bot needs a shared inbox, not a phone.
- 2. Build the knowledge base. Start with the twenty questions your team answers most, your price list, policies and product pages.
- 3. Define the agent. Name, role, writing voice, languages, what it may and may not discuss, and which CRM fields it should collect.
- 4. Set the handoff rules. Triggers, destinations, working hours and what happens on success and on failure.
- 5. Test with real conversations. Paste last month's actual customer messages and read the answers. Fix the knowledge base, not the prompt, wherever it guesses.
- 6. Launch on one inbox, then expand. Begin with WhatsApp or website chat, measure for two weeks, then add the other channels.
Measuring the results
Four metrics tell you whether the bot is working. Containment rate: the share of conversations resolved without a person. Handoff rate and reasons: where and why it escalates, which is the best guide to what to add to the knowledge base. Response and resolution time: usually seconds for the bot, and shorter for humans too because the details are already collected. Customer satisfaction: a rating at close, or an AI-scored sentiment. Review weekly for the first month and monthly afterwards.
AI Agents in Teloring
Teloring's AI Agents are configured in eight tabs (identity, writing voice, knowledge grounding, rules, the data to collect, what to do on success and on failure, and working hours) and appear in the assignment list next to the human team. They work on WhatsApp, website chat, Instagram, Messenger, Telegram and every other connected channel, answer from the Knowledge Base with a source reference, write into CRM fields, and hand off with a summary.
For the human side of the same conversations, AI Copilot suggests replies, detects mood, urgency, churn risk and upsell intent, and summarises each conversation on close. AI capabilities are switched on one at a time in AI World, are off by default and read only your own account's data. An enabled AI Agent counts as one seat on the pricing page.
Frequently asked questions
Yes, on the WhatsApp Business API. Meta requires that businesses identify themselves, respect opt-in for outbound messages and offer a way to reach a person. Bots on the consumer app or unofficial tools violate the terms and risk the number.
For a rule-based flow, yes: you design the branches. For an AI agent, you describe its role and rules, upload the content it should answer from and set the handoff conditions; it handles the phrasing itself.
By grounding: the agent answers only from approved content and says when it does not know. In Teloring you can require strict grounding and set a confidence threshold below which the conversation goes to a person.
Yes. AI agents detect the language of each message and reply in it, and you can restrict them to the languages your team supports.
In Teloring an enabled AI Agent counts as one seat, and AI actions and template messages are paid from monthly credits. There is no separate chatbot licence.